How does an LLM do spacial reasoning?
Reddit.com上的一位用户正在询问大型语言模型（LLM）如何进行空间推理，并提到了“Astra”在3D建模和游戏演示方面的能力。他们好奇这种能力是源于训练数据中学习到的模式、将自然语言指令转换为3D坐标系统的能力、熟练控制3D建模程序的能力，还是某种程序生成过程。该用户正在寻求教育资源，以了解其背后的机制。
- 发布
- 09/07 21:27 UTC+0
- 收录
- 09/08 00:00 UTC+0
- 来源类型
- 开发者社区
- 档位
- 社区
- 信源状态
- 正常
档位是按信源手工设定的编辑判断，不是逐条打分。
I’ve been seeing a lot of coverage of Astra’s spacial reasoning abilities, with 3d modelling demos and game demos.
Through what mechanism does an LLM “do” special training like this?
Is it through learned patterns of 3d modelling in training data?
Is it an ability to translate natural language instructions into 3d coordinate systems?
Is it really good at controlling 3d modelling programs and interfacing with them?
Is it a kind of procedural generation process, where is creating code that can do the job it’s trying to achieve?
I’d love to learn more about how this is being achieved and would appreciate any good educational links on this subject.
Edit: wanted to use the “question” flair, but it’s not available on mobile?